The banners and backdrops adorning Cisco’s recent strategy event might have read “Internet for the Future,” but the subtext was closer to “finally acknowledging the present.” The harsh reality of hyperscale cloud economics, fueled by component commodification and the rise of open (and increasingly open source) hardware and network software has finally pierced the enterprise IT bubble Cisco has comfortably occupied for decades. Cisco used the high-profile event featuring CEO Chuck Robbins and other execs to unveil a three-pronged strategy that encompasses new switch silicon and products using it, pluggable optical interfaces and software that the company hyperbolically states will build:
a new internet — one designed to push digital innovation beyond the performance, economic and power consumption limitations of current infrastructure. A multi-year approach that is defining the Internet for decades to come.
That’s right; Cisco’s strategy will reshape the internet for generationsto come. Outside the aura of Cisco PR, it looks more like a strategy designed to fend off technological dominance by merchant switch silicon providers like Broadcom and creeping irrelevance as cloud vendors and wireless carriers opt for in-house or open, white box network gear instead of Cisco’s pricey and proprietary hardware. Indeed, Cisco divulged precisely the necessary strategy to reflect a world where cloud service providers increasingly drive infrastructure hardware sales and product priorities and the innovative power of open systems using freely available hardware and software have diffused from servers to network equipment.
Cisco’s headline announcement is a new switching chip, Silicon One, with an internal design that Cisco claims is flexible enough to serve multiple roles, notably in standalone switches, line cards for chassis-based routers or in a distributed router fabric. Cisco has several configurations of the chip for products targeting different markets including carrier core and 5G mobile edge, enterprise and hyperscale distributed cloud networks. Such architectural commonality reduces product development costs for Cisco and product qualification and deployment testing by customers, although a Cisco-sponsored white paper overstates the architectural benefits in its TCO model by combining the savings from product consistency with the significant scaling and port consolidation benefits of higher throughput switches supporting 100G and 400G Ethernet interfaces.
Cisco’s claims the first generation Silicon One Q100 chip is “isthefirstroutingsilicontobreakthrough the10Tbpsbenchmarkfornetworkbandwidth,” which is patently false given that Broadcom has been shipping its Jericho2 programmable switch-router SoC in volume for almost a year. Cisco does leave itself an out for the ‘first ever’ claim by appending the following, “withoutcompromisingcarrier-class capabilities,e.g.,featurerichness,largequeueset,deepbuffers,largeNPUtables,and advancedprogrammability.” Those are the kind of subjective weasel words that sports statisticians like to dream up with claims of “the first baseball no-hitter by a left-hander shorter than six feet during a game a high altitude on a Tuesday.” Nonetheless, I’m sure Broadcom and its telco customers would quibble with any assertion that the Jericho2 isn’t “carrier-class.”
Specsmanship notwithstanding, a significant benefit of Cisco’s unified hardware architecture will be software consistency for the OS, patches and monitoring interfaces in a new XR7 version of Cisco’s network OS (NOS).
Cisco also announced the first product built on the Q100, the 8000 series switch routers available in both 1U and 2U fixed configurations with throughput of 10.8Tbps or as a modular chassis with up to 18 slots, each with 14.4 Tbps throughput,supporting both 100G and 400G ports. The products target carriers and cloud operators, i.e. the growth areas for network equipment given the increasing usage of cloud services and ongoing buildout of 5G infrastructure. The 8000 series will use a new IOS XR7 network OS that Cisco says is more efficient, uses less memory and supports hardware certificates to establish a root of trust for the OS and future upgrades.
Cisco also announced:
Aside from the technology introductions, Cisco made a significant change to its business model by announcing that it is disaggregating the main components of its new switch and allowing its optics, XR7 NOS and Silicon One switch chips to be purchased and used independently. Indeed, it is this decoupling of components, primarily the switch silicon, from Cisco’s packaged products that shows a company bending to market pressure.
The move also pits Silicon One against competitors from Broadcom, Intel — which recall, recently bought Barefoot Networks, a developer of programmable network processors — and several startups seeking to disrupt the Ethernet switch market. All are tapping into a trend towards switches built on openly available chips that was started by the hyperscale cloud operators, but increasingly embraced by enterprises and carriers opting for products from Arista, Juniper and various white box manufacturers. Indeed, IHS Markit estimates that products based on merchant silicon will account for 63 percent of the market in 2020.
The hyperscale operators’ motivation for using merchant silicon was its openness, offering them the ability to custom design switch hardware to their needs and run their NOS of choice such as Microsoft with SONiC and Facebook’s FBOSS. Carriers like AT&T have also jumped on the open networking bandwagon as they redesigned their infrastructure for 5G. Other IT organizations were attracted by the price-performance of white box hardware based on Broadcom’s leading edge technology. Indeed, as I mentioned above, previously released Broadcom chips like the Jericho2, Tomahawk 3 and Trident 4 boast specs as good or better than what Cisco just announced.
In Cisco’s latest earnings call, CEO Robbins states that the buildout of 5G infrastructure and the concomitant need of carriers for flexibility switching products, along with the cloud operators transition to 100G and 400G networks are critical to Cisco’s future growth and hence its product strategy. Hock Tan, CEO of Broadcom, Cisco’s biggest competitor in merchant switch silicon, discussed similar trends during his latest earnings call. In response to a question about cloud infrastructure spending and Cisco’s merchant silicon intentions, Tan sees renewed growth in cloud infrastructure spending in 2020 and welcomed the competition (emphasis added):
With respect to one of our very good customers, turning into -- coming into merchant silicon with the recent announcement I think yesterday on one silicon or the Silicon One in the router 8000, I think it's --we welcome that because it validates a couple of things we've been pushing for years. One of which is that they will be and there has been and will be more and more disaggregation of software, the operating system from hotlist, the silicon, the chips that supports it. … So the fact that Cisco has joined it, now they validate the model, the trend we have been pushing and it's great to see that we're right in that regard.So we will welcome the competition.I just wanted to add. It's more than cloud that we're seeing that happen. It's also in enterprise, traditional enterprise, particularly some of the large telcos, who you classify as cloud too. … Jericho 2 router [chip] , which we're using today to enable the path of those telcos [and] enterprises toward disaggregation of hardware and software has been around, and theyhave been shipping it for over a year, and that runs 10 terabit per second, the same bandwidth as Silicon One announced yesterday. And earlier this week, we announced as a future successor 25.6 terabit per second switching and routing.
Tan also noted Broadcom’s broad portfolio, which spans carrier routing to low end campus switching, using the same APIs, a goal that inspired Cisco to develop the Silicon One architecture.
Disaggregating its router/switch products via Silicon One and its support for open hardware standards like SONiC puts Cisco on a more competitive path and better positions it to capitalize on the fastest growing areas in networking equipment: infrastructure for carrier 5G networks, hyperscale cloud operators and global online services and social networks. The market agreed, with Cisco’s stock popping three percent after the announcements. While it is arguably playing catch-up to Broadcom and its customers like Arista, Cisco has the resources to excel in these markets and there’s ample time to win new business outside its traditional strength in enterprise IT.
As with databases, enterprises are historically resistant to changing network equipment vendors, however, Cisco’s openness to merchant silicon, unbundled hardware and alternative NOSs will nonetheless yield benefits. Enterprises should see more aggressive pricing and access to new innovative features as Cisco is forced to respond to hardware competition from Broadcom and Intel and innovators in network software like Arccus, Cumulus, Pica8 and open source projects like SONiC.